Jasper Kars: "As a society, we sometimes need to say 'no, we won’t be using this AI model’"

When Jasper Kars started working at the Ministry of the Interior and Kingdom Relations five years ago, AI was “still a niche field, certainly within policy”, he explains as he receives DUB in the historic building housing the Department of Public Administration and Organisation Studies on Bijlhouwerstraat.

Even then, however, AI caught his interest. “While studying Political Science and Data Science, I became interested in how large bureaucracies grapple with new technologies. They want to regulate them, but they also want to use those technologies to accomplish their own tasks and improve their operations.”

ChatGPT was launched at the end of 2022, making generative AI accessible to a wider audience. About a year later, Kars was given the opportunity to pursue a PhD alongside his job. “It quickly dawned on me that I wanted to research generative AI. At that point, it was all new, but I knew it would be hugely significant."

“GenAI is radically changing the speed, scale and way in which we process information. Free information and open debate are the cornerstones of our democracy. At a time when both are under pressure in many parts of the world, this is a real test.”

Interaction
Kars is investigating how AI is changing democracy and how democracy is trying to steer AI development. He is examining four areas: the structure of our system of government, politics, how this translates into policy, and the social and political debate.

According to him, AI and democracy can “clash or reinforce one another”. He believes that autonomous AI systems – which are capable of taking over entire government processes – pose an ever-greater risk to democracy, as it becomes difficult to trace how democratic decisions were reached.

However, he also believes that the government can use AI to improve its services and strengthen democracy. Examples include processing appeals or Freedom of Information requests more quickly, and drafting clearer letters at different language levels.

“I do not believe in technological determinism, according to which technology is the driving force behind social change. It is the interplay between technology and social factors that determines how a technology ultimately finds its way into society.”

Politics
Research by the University of Amsterdam shows that political parties used generative AI extensively during this year’s municipal elections. They spread AI-generated images, including deepfakes and infographics, as part of their campaigns.

Political parties in the Netherlands also use generative AI when drafting motions. “We know for certain that at least one motion was drafted using generative AI, as the proposer was transparent about that. It is likely that others are also using AI to draft motions, ask questions or write letters to Parliament, but that is harder to say, as parties are not transparent about it.”

He says it is problematic when AI takes over a big chunk of the democratic process. “For example, when a random local councillor submits a huge number of questions, which have been drafted in the blink of an eye with the help of AI. This can put the administration and the civil service under severe pressure. In that case, it is no longer the elected representative who is carrying out the actual political work, but an AI system.”

Taking over processes
Kars also notes that the government, including executive bodies and local authorities, is increasingly experimenting with AI. For example, government organisations are using chatbots for their initial contacts with members of the public. “In addition, local authorities are using AI to identify people who are not being reached by certain social benefits, even though they are entitled to them.”

These are relatively straightforward applications, but according to Kars, certain governmental uses of AI “can have major implications for democracy”. This rings true for autonomous AI systems, also known as agentic AI, which can take over entire processes.

The Dutch government is also experimenting with this type of AI, with the Tax and Customs Administration recently conducting a study of its possibilities. According to Kars, the government must ensure that “processes remain properly traceable and that it does not become dependent on systems.”

Discrimination by AI
As AI systems become more autonomous, traceability becomes harder, which poses a threat to our democracy. “A fundamental requirement of democracy is that citizens, journalists and researchers are able to obtain information about government decisions,” says Kars. “If decisions are no longer transparent, citizens cannot properly assert their rights.”

However, things can also go wrong when simpler algorithms are used, as illustrated by the Dutch childcare benefits scandal and the case of students investigated for fraud in student financing schemes. In both cases, citizens with migrant backgrounds were more likely to be investigated. “We see that when AI is misused, minorities are often the ones who suffer, even though the government must not be biased. All citizens are equal before the law. Otherwise, important democratic values such as justice, equal treatment and participation are jeopardised.”

Well-informed citizens 
Furthermore, according to Kars, the rise of generative AI is having a major impact on government communications. “We are seeing a decline in the number of visitors to government websites. Thanks to chatbots, people are finding information in different ways and no longer need to visit the official government website to get an answer to their question. As a result, governments become largely dependent on major tech companies that control information channels such as social media and AI chatbots to communicate online with citizens."

According to Kars, it is important for the government to reach its citizens directly, rather than through AI systems. “If the way information is provided fundamentally changes, that also has major implications for a democracy,” he says. “A healthy democracy requires people to be able to participate in an informed manner, such as by voting and taking part in civic engagement or the public debate. To make a well-considered choice, citizens rely on the news media, political parties and the government itself.”

Social media
Public debate has also shifted to social media. “You could say that the internet was a blessing for open and free debate, because everyone suddenly had the chance to make their voice heard. The reality is that social media’s algorithms and AI determine what is and isn’t shown, with sensational and extreme content often taking precedence. That plays right into the hands of the fringes with their extreme views.”

In his view, social media companies are not taking sufficient responsibility. “There is so much disinformation and misinformation that does not get blocked by social media companies’ moderation teams. Generative AI acts as a kind of performance-enhancing drug, as it makes it very easy to create disinformation and launch a disinformation campaign, whilst further eroding trust in the democratic process,” ponders Kars. “During the recent parliamentary elections, we saw an increase in the spread of AI-generated misinformation and disinformation intended to discredit political opponents. This is very worrying.”

Opportunities
Kars does not only identify threats to democracy. In his view, if AI is used correctly, it can also strengthen democracy. The government could use AI to improve its services.

Since AI can process considerable amounts of information quickly, it could assist with policy-making and serve as a tool in coalition negotiations. It could also compare new policies with those adopted previously. “We could make better use of the government’s extensive knowledge,” says Kars. “When negotiating coalition agreements, parties could use AI to identify key documents and existing agreements relating to a specific policy issue, as well as lay the groundwork for reaching consensus, and put forward new proposals if negotiations threaten to stall.”

Furthermore, politicians can use AI to arrive at more nuanced insights. “If you only read big Dutch newspapers such as De Volkskrant or De Telegraaf, your information is limited. AI can gather insights from all sorts of online sources – not just the best-known ones. Public debate is increasingly taking place online, and alternative media are becoming more popular. Politicians need to stay in touch with society, and AI can give them a better picture of that broad spectrum of opinions. At the same time, this requires politicians to be even more critical in their use of sources, as online information can also be incorrect, misleading, or deliberately manipulated — something that has only become easier with the rise of AI.”

AI regulation
According to Kars, for a government or other democratic institution to deploy AI, certain conditions must be met, and safeguards must be in place in case AI goes off the rails. Therefore, a good AI tool is essential. Kars: “AI is not neutral. Although it is impossible to make AI completely neutral, an AI model that could affect wider groups in society must not be based on a one-sided worldview."

“The AI model must be trained according to the values of a particular community or society. In addition, the use of AI must be lawful. Proper oversight – both internally, within the organisation using it, and by independent regulators – is a key prerequisite for this.”

According to Kars, society and politics must debate how we want to handle AI. As long as there is no clear answer to this question and the conditions for safe AI have not yet been widely embraced, he advocates continuing to monitor AI's development critically. “As a society, we must sometimes say: 'No, this AI model is not yet ready for use; no, it is potentially too disruptive.' However, this assessment must lie with elected representatives, not with the big tech companies developing AI. Such balance is still lacking in the AI era.”